The Agent Skills Directory

[SAFE]: The skill uses dynamic context injection via shell commands to locate configuration files within the project structure. These commands are hardcoded, restricted to the .claude/agents directory, and do not involve network access or sensitive file exposure.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests and processes data from external agent configuration files. While this data could contain malicious instructions, the skill's limited toolset and read-only nature minimize potential risks.
Ingestion points: Agent configuration files read during the discovery phase.
Boundary markers: No explicit markers or warnings are used when processing the content of identified files.
Capability inventory: The skill uses Glob, Read, and TodoWrite for analysis and reporting. It has no capabilities for code execution or network operations.
Sanitization: There is no evidence of sanitization or validation of the ingested content before it is processed for the audit report.

meta-audit